This project is intended to be a forensic investigation tool. It enables extraction of key frequency ranges from audio samples. The background signals can be used to compare multiple different sources or audio files and determine if they were recorded at the same place or at the same time.
The electrical grid in the US runs at 60Hz with very little deviation. Historical data of the exact signal produced along the grid can be accessed by investigators. By extracting a narrow frequency band around 60Hz in audio recordings taken as evidence, it is theoretically possible to compare the ~60Hz component of the audio evidence to historical readings of the grid to determine an exact time when a file was recorded. This information is extremely valuable in instances where metadata for a file was overwritten or otherwise modified.
For more information on how this project works, see the detailed write up at Mains Hum and Audio Evidence Comparison Tool.pdf.
Requires:
- python: It works on python3.11.2, but will probably work on many other versions
- ffmpeg: ffmpeg is a command line audio and video editor. The official download page can be found here: https://ffmpeg.org/download.html
Run pip install -r requirements.txt to install any additional needed libraries.
This project comes with a cli, which is the primary way of interfacing with it. There are two tasks that can be accomplished through the cli, explained below.
Real world investigations focusing on comparing evidence against the mains hum would not need this mode, but in cases where an investigator does not have access to historical mains data or would like to compare audio evidence on different frequency ranges, this mode is useful.
This mode takes one sample file and extracts a given range of frequencies from it, writing the extracted data into a new .wav file.
python cli.py x [sample file] [frequency] [range] [output file]
- x: Specifies extract mode. Can also be written as
extract. - sample file: Takes a relative path for the file from which data should be extracted. Can be any file that contains audio data.
- frequency: The core frequency in Hz that should be extracted from the sample.
- range: How many Hz in either direction from the core frequency should be accepted through the filter. This is not an exact measure due to how the filtering method works.
- output file: Where the extracted signal should be saved to. Should end in
.wav.
After running extract, a new file will be created which contains only the desired frequencies from the input signal.
This is the primary mode of operation of the tool. Compare mode takes two files, one being an unedited, unprocessed sample to be analyzed. The other is a known background signal, generally in a very narrow frequency band that can be used to time stamp the sample. The program will return if certain background frequencies of the sample also exist within the background, and if so, when.
python cli.py c [sample file] [background data] [freq] [range]
- c: Specifies compare mode. Can also be written as
compare. - sample file: Takes a relative path for the file from which data should be analyzed. Can be any file that contains audio data.
- background data: Takes a relative path for the file which contains historical background data with known recording time. Must be a .wav file.
- frequency: The core frequency in Hz that should be extracted from the sample.
- range: How many Hz in either direction from the core frequency should be accepted through the filter. This is not an exact measure due to how the filtering method works.
The frequency range specified should be roughly equivalent to the expected frequency range of the background. For the US power grid, the frequency range of that signal is expected to be 60Hz +/- 1%1. This means a reasonable frequency range for comparing against the background mains hum would be 60Hz +/- 0.6 Hz.
Prints whether a match was found between the sample and the background, and at what sample point in the background file does the match start.
When extracting narrow frequency bands, the resulting audio file is extremely quiet. There is a function in the code that attempts to normalize output files to audible levels, but it is imperfect, and may cause issues during comparison. For most samples, the normalization function produces viable audible output, but in others it causes extreme distortion. There is no interface to enable this feature. Line 22 of audioAnalyzer.py should be uncommented, and line 23 should be commented out to enable this feature.
The cli expects two arguments that specify which frequencies should be extracted. The first is the average frequency to be extracted, and the second is the distance from the average that is acceptable to be extracted. It may be more intuitive in some use cases to instead specify the arguments as minimum frequency and maximum frequency, and it is an easy mistake to make. For the intended use case of this tool, the implmented interface is the best way to specify which frequencies to extract.
The comparison method in this code is extremely brittle against any perturbations in the signal between the sample and the background signal. This is a known limitation of the tool, and a full explanation can be found in Mains Hum and Audio Evidence Comparison Tool.pdf
The compare function only performs frequency extraction on the sample file. The background file is expected to already be a narrow frequency band signal. If this is not the case, you must first run the background audio through the extract mode, and then use that output as the background signal in the compare mode.
Footnotes
-
Muelaner, J. (2021, February 5). Grid frequency stability and renewable power. Engineering.com. https://www.engineering.com/grid-frequency-stability-and-renewable-power/ ↩